From the course: Security Risks in AI and Machine Learning: Categorizing Attacks and Failure Modes
Unlock this course with a free trial
Join today to access over 24,900 courses taught by industry experts.
Backdoors and existing exploits
From the course: Security Risks in AI and Machine Learning: Categorizing Attacks and Failure Modes
Backdoors and existing exploits
- [Instructor] If you've ever played a video game, you may have used a cheat code, a special combination of button presses that give you an advantage. For example, a sequence like up, up, down, down, left, right, left, right, B, A might give you extra lives, invincibility, or enable diagnostic mode. A code like this, while harmless in games, can be a serious security risk in other contexts. Anyone who's seen the classic movie, "War Games" knows that the developers of non-gaming applications can leave backdoors in their systems to allow someone knowing the code to gain access or special privileges. backdoors rely on a trigger, a special input that brings about specific response. This trigger could be a specific prompt sent to an LLM coding assistant that causes it to include malware in the code it generates, or it might be a specific visual signal, such as a sticker on a road sign that causes a self-driving car to…
Contents
-
-
-
-
(Locked)
Perturbation attacks and malicious input7m 26s
-
(Locked)
Poisoning attacks6m 54s
-
(Locked)
Reprogramming3m 30s
-
(Locked)
Physical domain: 3D adversarial objects3m 55s
-
(Locked)
Supply chain attacks4m 4s
-
(Locked)
Model inversion5m 19s
-
(Locked)
System manipulation4m 49s
-
(Locked)
Membership inference and model stealing4m 26s
-
(Locked)
Backdoors and existing exploits3m 45s
-
(Locked)
-
-
-